Research And Application Of Decision-making Model For Video Games Based On Deep Reinforcement Learning

With the continuous development of the Internet and the continued popularity of various electronic devices such as mobile phones,games have become a very important part of people's lives.People know different people through games.Games have also become a pastime for people to rest.The game industry has also witnessed rapid development.How to bring differentiating game experiences to game players has become a top priority.With the continuous research and development of machine learning,deep reinforcement learning based on deep learning and reinforcement learning has provided a solution for researchers using machine learning in games.This paper starts with deep reinforcement learning to study how machine learning is applied in games.The main tasks are as follows:(1)This paper introduces the development status of deep reinforcement learning at home and abroad,collates the deep reinforcement learning algorithms used in the game,and analyzes the principle of these algorithms,and analyzes the core technologies.(2)Research deep learning and reinforcement learning.Reinforcement learning is an effective method to solve exploration and decision problems.However,reinforcement learning is challenging when dealing with high-dimensional data.Therefore,neural networks used in deep learning are used to process high-dimensional sensing data(eg,video,speech,etc.)Extract features as input for reinforcement learning.The deep convolutional neural network has a natural advantage in processing images.The network is very good at detecting features,but the detection efficiency of the feature variants is not so good.Therefore,the capsule network is used instead of the convolutional network to process the sensing data.Q-learning is a classic reinforcement learning method.Its core is to approach the optimal strategy step by step through continuous interaction with the environment,trial and error,and feedback.(3)Research deep reinforcement learning.Google's artificial intelligence team proposed a deep reinforcement learning algorithm that combines deep learning and reinforcement learning.It has been successfully applied in the Atari game environment and has achieved amazing results.However,this algorithm has the problem of inefficient exploration.This article uses the improved Bootstrapped DQN method to improve the efficiency of exploration.(4)Design a video game decision model that combines the capsule network andBootstrapped DQN method.By using a capsule network instead of a convolutional neural network,the game decision model is improved in its ability to detect feature variants;the Bootstrapped DQN method is used to improve the game decision-making model's training speed and depth exploration capabilities.This article designs a video game decision model that combines the capsule network with Bootstrapped DQN method.The game decision model uses the capsule network to perceive high-dimensional input data and extract features.The capsule network can not only detect features,but also learn feature deformation and reduce the error rate when extracting features.In addition,the Bootstrapped DQN method is used for deep exploration strategies to speed up the efficiency of model exploration strategies.Through experimental analysis and comparison,this game decision model can effectively learn the control strategy and improve the efficiency of the exploration strategy.